(Invited) Chemical Mechanical Planarization Integrated Data Collaboration: Uncover Hidden Interactions for Predictive Maintenance and PCN

Abstract

This article introduces an innovative analytical methodology that employs machine learning to optimize and predict consumable impacts for Chemical Mechanical Planarization (CMP). By harnessing fleet CMP equipment performance and individual equipment characteristics, our proposed approach aims to elevate wafer-level quality, maximize equipment utilization rates, and minimize unnecessary downtime. CMP, a multifaceted process, its performance is often influenced by strong interactions from its critical components including slurry, pad, machine recipe, and dresser usage. This requires control beyond hardware advancements, especially when equipment is pushed to its limits. Therefore, the demand for higher precision on-wafer performance mandates a data collaboration with sophisticated analytical strategy. In addition, a digitalized product change notification (PCN) concept is also introduced to demonstrate effectiveness of applying integrated data analytics collaboratively across semiconductor chain to accelerate material manufacturing process changes, minimizing impacts on wafer results and shortening the learning cycle and ramp time to full production.Our study leverages data analytics to explore differences between equipment within the same fleet of similar processing. Engineers can deploy a process based on the historical performance of specific equipment, supported by a model trained on data from the same fleet. Furthermore, our approach enables each equipment to predict its own fingerprinted consumable maintenance, accurately forecasting the maintenance time for consumable changes. This concept is equally applicable for consumable PCN migrations.The first use case is to demonstrate the use of integrated data for fleet matching of CMP equipment and/or consumable batches running similar processes. This is particularly crucial as the demand for process precision increases and the need to reduce post-CMP result variability in removal rates and wafer-level uniformity intensifies. This approach could add proactive data driven strategy for process engineers who traditionally concentrate on the removal rate, a highly influential factor, with the standard deviation of platen and machine data identified as a key parameter.The second use case employs heatmaps analysis and proposes scatter plots as an intuitive method to comprehend intricate connections between average removal rates and various parameters. Machine learning is particularly advantageous in this context, as no single parameter plays a dominant role. Preliminary observations suggested, for example that dresser and pad characteristics may have compounding impacts to the wafer level performance.In summary, this article provides a roadmap for in-depth CMP data analysis fed into machine learning, aiming to unravel trends and influential factors that can enhance process optimization and shorten the PCN qualification timeline to meet the increasing demands for CMP process precision and efficiency in High Volume Manufacturing (HVM).